On-Device & Local InferenceTim Dettmers et al.
Dettmers et al. present LLM.int8()
Presented LLM.int8() — converts 175B checkpoints to 8-bit without performance degradation, halving inference memory
Why it matters
Made it possible to run large models on consumer-grade GPU servers
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2208.07339Added in Week 35, 2026 · See that week